Swarm Tree Orchestration Squad

Recursive tree decomposition — plans the COMPLETE tree before executing.

2-phase pipeline: full planning → wave-based execution via TaskCreate, with UxIxR scoring, circuit breaker, and automatic integration.

/sto

The Problem with Complex Tasks

You know what happens when an agent tries to execute a huge task directly.

It starts, loses track, produces incomplete code, or simply stalls halfway.

The cause: Without prior planning, the agent doesn't know the real scope, doesn't manage dependencies between sub-tasks, and wastes context.

The Swarm Tree Orchestration Squad solves this with a simple rule:

Plan the COMPLETE tree before executing any leaf.

Before and After

How It Works

┌────────────────────────────────────────────────┐
│              PHASE 1 — PLANNING                │
│                                                │
│  P0: Setup .swarm-tree/                        │
│  P1: TaskDecomposer → rawTree (recursive)      │
│  P2: NodeScorer → UxIxR scoring + dep graph    │
│  P3: TreeOrchestrator → tree.json + context    │
│  P4: ContextComposer → prompts per node        │
│  P5: TreeValidator → WBS + circuit breaker     │
│  P6: Present for user validation               │
└───────────────────────┬────────────────────────┘
                        │ user validates
                        ▼
┌────────────────────────────────────────────────┐
│              PHASE 2 — EXECUTION               │
│                                                │
│  WaveDispatcher → parallel waves via           │
│  TaskCreate (respecting dependencies)          │
│                    │                           │
│  Unified polling + reactive dispatch           │
│                    │                           │
│  TreeIntegrator → validate artifacts +         │
│  fix imports + generate final report           │
└────────────────────────────────────────────────┘

Agents

Commands

/sto *orchestrate-tree TASK="..."

/sto *plan-tree TASK="..."

/sto *resume-tree

/sto *analyze-tree